Imagine transforming a group of ordinary Macs into a powerhouse AI supercomputer with the upcoming macOS Tahoe 26.2 update. But wait, isn't that what the Mac Pro is for? Well, Apple has a new trick up its sleeve.
The Power of Mac Clusters: With the new low-latency feature, you can link multiple Macs using Thunderbolt 5, creating a unified computing beast. This setup is a game-changer for developers and researchers who need to run resource-intensive AI models locally. By connecting four Mac Studios, each capable of handling 512GB of unified memory, you can execute the 1 trillion parameter Kimi-K2-Thinking model with unprecedented efficiency, leaving power-draining GPU-based PCs in the dust.
But here's where it gets even more intriguing: Thunderbolt Evolution: Previous Mac clusters were constrained by slower Thunderbolt speeds, especially when hubs were involved, limiting data transfer rates to 10 Gb/s. However, Apple's new feature unleashes the full potential of Thunderbolt 5, enabling speeds of up to 80 Gb/s. And it's not just the Mac Studio that can join the party; the M4 Pro Mac mini and M4 Pro/Max MacBook Pro are also compatible, allowing developers to build clusters without breaking the bank on specialized hardware.
I witnessed the magic in action as a cluster of four Mac Studios effortlessly loaded and ran the massive Kimi-K2-Thinking model on an early version of ExoLabs's EXO 1.0. The most impressive part? This cluster consumed less than 500 watts of power, a mere fraction of the energy required by typical GPU clusters, like NVIDIA's RTX 5090, which demands 575W and often more.
The MLX Boost: macOS Tahoe 26.2 also brings a significant enhancement to Apple's open-source MLX project, granting full access to the neural accelerators on the M5 chip, promising faster AI inferencing. Ironically, the only M5 Mac available, the 14-inch MacBook Pro, supports Thunderbolt 4, making it incompatible with the new Mac clustering feature.
Apple Silicon's unified memory and energy efficiency already made Macs appealing for AI tasks, but the ability to cluster systems over Thunderbolt 5 takes it to the next level. While a Mac Studio with 512GB of RAM is an investment (starting at $9,499 with the M3 Ultra chip), it's not the only option. Labs and businesses with existing Mac Studios, Mac minis, and MacBook Pros can now consider clustering their devices to unlock new levels of AI performance.
